Verification Before Completion

This is a guard rail that forces you to actually run verification commands before claiming anything is done. It blocks the "looks good, tests should pass" reflex and requires fresh evidence for every completion claim. The five step pattern is simple: identify the verification command, run it completely, read the full output, verify it matches your claim, then and only then say you're done. The skill cites real numbers showing unverified work costs 60x more to fix than just running the test upfront. It's blunt about the core issue: claiming success without proof isn't efficiency, it's dishonesty. Use it as a pre-commit gate or whenever you're about to mark something complete.

Install to Claude Code

npx -y skills add bobmatnyc/claude-mpm-skills --skill verification-before-completion --agent claude-code